🔄 Using Different Models
To use a different trained MAPPO model:
- Locate your checkpoint:
Results/mappo_sumo_v4/seed_<n>/models/<run_name>/<steps>/agent.th
Example:
- Results/mappo_sumo_v4/seed_42/models/.../2828160/agent.th
- Copy to trained_models directory:
cp "<source_path>/agent.th" "rl-inference-service/app/trained_models/agent.th"
- Rebuild and restart:
docker build -t rl-inference-service:2.0.0 ./rl-inference-service
docker compose up -d
Note: The model is an EPyMARL RNNAgent checkpoint (PyTorch
.thfile), not a Stable-Baselines3.zip. Architecture:fc1(24→128) → GRUCell(128→128) → fc2(128→4). Input shape = 24 (19 obs + 5 agent-id one-hot).